VM: only single page chunks
[minix.git] / sys / ufs / chfs / chfs_gc.c
blobaa32d64b96760a8d4eebfd90d07daed1e2a341e6
1 /* $NetBSD: chfs_gc.c,v 1.2 2011/11/24 21:09:37 agc Exp $ */
3 /*-
4 * Copyright (c) 2010 Department of Software Engineering,
5 * University of Szeged, Hungary
6 * Copyright (c) 2010 Tamas Toth <ttoth@inf.u-szeged.hu>
7 * Copyright (c) 2010 Adam Hoka <ahoka@NetBSD.org>
8 * All rights reserved.
10 * This code is derived from software contributed to The NetBSD Foundation
11 * by the Department of Software Engineering, University of Szeged, Hungary
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
35 #include "chfs.h"
37 void chfs_gc_release_inode(struct chfs_mount *,
38 struct chfs_inode *);
39 struct chfs_inode *chfs_gc_fetch_inode(struct chfs_mount *,
40 ino_t, uint32_t);
41 int chfs_check(struct chfs_mount *, struct chfs_vnode_cache *);
42 void chfs_clear_inode(struct chfs_mount *, struct chfs_inode *);
45 struct chfs_eraseblock *find_gc_block(struct chfs_mount *);
46 int chfs_gcollect_pristine(struct chfs_mount *,
47 struct chfs_eraseblock *,
48 struct chfs_vnode_cache *, struct chfs_node_ref *);
49 int chfs_gcollect_live(struct chfs_mount *,
50 struct chfs_eraseblock *, struct chfs_node_ref *,
51 struct chfs_inode *);
52 int chfs_gcollect_vnode(struct chfs_mount *, struct chfs_inode *);
53 int chfs_gcollect_dirent(struct chfs_mount *,
54 struct chfs_eraseblock *, struct chfs_inode *,
55 struct chfs_dirent *);
56 int chfs_gcollect_deletion_dirent(struct chfs_mount *,
57 struct chfs_eraseblock *, struct chfs_inode *,
58 struct chfs_dirent *);
59 int chfs_gcollect_dnode(struct chfs_mount *,
60 struct chfs_eraseblock *, struct chfs_inode *,
61 struct chfs_full_dnode *, uint32_t, uint32_t);
63 /* must be called with chm_lock_mountfields held */
64 void
65 chfs_gc_trigger(struct chfs_mount *chmp)
67 struct garbage_collector_thread *gc = &chmp->chm_gc_thread;
69 //mutex_enter(&chmp->chm_lock_sizes);
70 if (gc->gcth_running &&
71 chfs_gc_thread_should_wake(chmp)) {
72 cv_signal(&gc->gcth_wakeup);
74 //mutex_exit(&chmp->chm_lock_sizes);
78 void
79 chfs_gc_thread(void *data)
81 struct chfs_mount *chmp = data;
82 struct garbage_collector_thread *gc = &chmp->chm_gc_thread;
84 dbg_gc("[GC THREAD] thread started\n");
86 mutex_enter(&chmp->chm_lock_mountfields);
87 while (gc->gcth_running) {
88 /* we must call chfs_gc_thread_should_wake with chm_lock_mountfields
89 * held, which is a bit awkwardly done here, but we cant relly
90 * do it otherway with the current design...
92 if (chfs_gc_thread_should_wake(chmp)) {
93 // mutex_exit(&chmp->chm_lock_mountfields);
94 if (chfs_gcollect_pass(chmp) == ENOSPC) {
95 dbg_gc("No space for garbage collection\n");
96 panic("No space for garbage collection\n");
97 /* XXX why break here? i have added a panic
98 * here to see if it gets triggered -ahoka
100 break;
102 /* XXX gcollect_pass drops the mutex */
103 mutex_enter(&chmp->chm_lock_mountfields);
106 cv_timedwait_sig(&gc->gcth_wakeup,
107 &chmp->chm_lock_mountfields, mstohz(100));
109 mutex_exit(&chmp->chm_lock_mountfields);
111 dbg_gc("[GC THREAD] thread stopped\n");
112 kthread_exit(0);
115 void
116 chfs_gc_thread_start(struct chfs_mount *chmp)
118 struct garbage_collector_thread *gc = &chmp->chm_gc_thread;
120 cv_init(&gc->gcth_wakeup, "chfsgccv");
122 gc->gcth_running = true;
123 kthread_create(PRI_NONE, /*KTHREAD_MPSAFE |*/ KTHREAD_MUSTJOIN,
124 NULL, chfs_gc_thread, chmp, &gc->gcth_thread,
125 "chfsgcth");
128 void
129 chfs_gc_thread_stop(struct chfs_mount *chmp)
131 struct garbage_collector_thread *gc = &chmp->chm_gc_thread;
133 /* check if it is actually running. if not, do nothing */
134 if (gc->gcth_running) {
135 gc->gcth_running = false;
136 } else {
137 return;
139 cv_signal(&gc->gcth_wakeup);
140 dbg_gc("[GC THREAD] stop signal sent\n");
142 kthread_join(gc->gcth_thread);
143 #ifdef BROKEN_KTH_JOIN
144 kpause("chfsthjoin", false, mstohz(1000), NULL);
145 #endif
147 cv_destroy(&gc->gcth_wakeup);
150 /* must be called with chm_lock_mountfields held */
152 chfs_gc_thread_should_wake(struct chfs_mount *chmp)
154 int nr_very_dirty = 0;
155 struct chfs_eraseblock *cheb;
156 uint32_t dirty;
158 KASSERT(mutex_owned(&chmp->chm_lock_mountfields));
160 if (!TAILQ_EMPTY(&chmp->chm_erase_pending_queue)) {
161 dbg_gc("erase_pending\n");
162 return 1;
165 if (chmp->chm_unchecked_size) {
166 dbg_gc("unchecked\n");
167 return 1;
170 dirty = chmp->chm_dirty_size - chmp->chm_nr_erasable_blocks *
171 chmp->chm_ebh->eb_size;
173 if (chmp->chm_nr_free_blocks + chmp->chm_nr_erasable_blocks <
174 chmp->chm_resv_blocks_gctrigger && (dirty > chmp->chm_nospc_dirty)) {
175 dbg_gc("free: %d + erasable: %d < resv: %d\n",
176 chmp->chm_nr_free_blocks, chmp->chm_nr_erasable_blocks,
177 chmp->chm_resv_blocks_gctrigger);
178 dbg_gc("dirty: %d > nospc_dirty: %d\n",
179 dirty, chmp->chm_nospc_dirty);
181 return 1;
184 TAILQ_FOREACH(cheb, &chmp->chm_very_dirty_queue, queue) {
185 nr_very_dirty++;
186 if (nr_very_dirty == chmp->chm_vdirty_blocks_gctrigger) {
187 dbg_gc("nr_very_dirty\n");
188 return 1;
192 return 0;
195 void
196 chfs_gc_release_inode(struct chfs_mount *chmp,
197 struct chfs_inode *ip)
199 dbg_gc("release inode\n");
200 //mutex_exit(&ip->inode_lock);
201 //vput(ITOV(ip));
204 struct chfs_inode *
205 chfs_gc_fetch_inode(struct chfs_mount *chmp, ino_t vno,
206 uint32_t unlinked)
208 struct vnode *vp = NULL;
209 struct chfs_vnode_cache *vc;
210 struct chfs_inode *ip;
211 dbg_gc("fetch inode %llu\n", (unsigned long long)vno);
213 if (unlinked) {
214 dbg_gc("unlinked\n");
215 vp = chfs_vnode_lookup(chmp, vno);
216 if (!vp) {
217 mutex_enter(&chmp->chm_lock_vnocache);
218 vc = chfs_vnode_cache_get(chmp, vno);
219 if (!vc) {
220 mutex_exit(&chmp->chm_lock_vnocache);
221 return NULL;
223 if (vc->state != VNO_STATE_CHECKEDABSENT) {
224 //sleep_on_spinunlock(&chmp->chm_lock_vnocache);
225 mutex_exit(&chmp->chm_lock_vnocache);
226 /* XXX why do we need the delay here?! */
227 // kpause("chvncabs", true, mstohz(50), NULL);
228 KASSERT(mutex_owned(&chmp->chm_lock_mountfields));
229 cv_timedwait_sig(
230 &chmp->chm_gc_thread.gcth_wakeup,
231 &chmp->chm_lock_mountfields, mstohz(50));
233 // KASSERT(!mutex_owned(&chmp->chm_lock_vnocache));
234 } else {
235 mutex_exit(&chmp->chm_lock_vnocache);
237 return NULL;
239 } else {
240 dbg_gc("vnode lookup\n");
241 vp = chfs_vnode_lookup(chmp, vno);
242 //VFS_VGET(chmp->chm_fsmp, vno, &vp);
244 dbg_gc("vp to ip\n");
245 ip = VTOI(vp);
246 KASSERT(ip);
247 //mutex_enter(&ip->inode_lock);
249 return ip;
252 extern rb_tree_ops_t frag_rbtree_ops;
255 chfs_check(struct chfs_mount *chmp, struct chfs_vnode_cache *chvc)
257 struct chfs_inode *ip;
258 struct vnode *vp;
259 int ret;
261 ip = pool_get(&chfs_inode_pool, PR_WAITOK);
262 if (!ip) {
263 return ENOMEM;
266 vp = kmem_zalloc(sizeof(struct vnode), KM_SLEEP);
268 ip->chvc = chvc;
269 ip->vp = vp;
271 vp->v_data = ip;
273 rb_tree_init(&ip->fragtree, &frag_rbtree_ops);
274 TAILQ_INIT(&ip->dents);
276 ret = chfs_read_inode_internal(chmp, ip);
277 if (!ret) {
278 chfs_clear_inode(chmp, ip);
281 pool_put(&chfs_inode_pool, ip);
283 return ret;
286 void
287 chfs_clear_inode(struct chfs_mount *chmp, struct chfs_inode *ip)
289 struct chfs_dirent *fd, *tmpfd;
290 struct chfs_vnode_cache *chvc;
293 /* XXX not sure if this is the correct locking */
294 // mutex_enter(&chmp->chm_lock_vnocache);
295 chvc = ip->chvc;
296 /* shouldnt this be: */
297 //bool deleted = (chvc && !(chvc->pvno || chvc->nlink));
298 int deleted = (chvc && !(chvc->pvno | chvc->nlink));
300 if (chvc && chvc->state != VNO_STATE_CHECKING) {
301 // chfs_vnode_cache_state_set(chmp, chvc, VNO_STATE_CLEARING);
302 chvc->state = VNO_STATE_CLEARING;
305 if (chvc->v && ((struct chfs_vnode_cache *)chvc->v != chvc)) {
306 if (deleted)
307 chfs_mark_node_obsolete(chmp, chvc->v);
308 //chfs_free_refblock(chvc->v);
310 // mutex_enter(&chmp->chm_lock_vnocache);
312 chfs_kill_fragtree(&ip->fragtree);
314 fd = TAILQ_FIRST(&ip->dents);
315 while (fd) {
316 TAILQ_REMOVE(&ip->dents, fd, fds);
317 chfs_free_dirent(fd);
318 fd = TAILQ_FIRST(&ip->dents);
322 TAILQ_FOREACH_SAFE(fd, &ip->dents, fds, tmpfd) {
323 chfs_free_dirent(fd);
326 if (chvc && chvc->state == VNO_STATE_CHECKING) {
327 chfs_vnode_cache_set_state(chmp,
328 chvc, VNO_STATE_CHECKEDABSENT);
329 if ((struct chfs_vnode_cache *)chvc->v == chvc &&
330 (struct chfs_vnode_cache *)chvc->dirents == chvc &&
331 (struct chfs_vnode_cache *)chvc->dnode == chvc)
332 chfs_vnode_cache_remove(chmp, chvc);
337 struct chfs_eraseblock *
338 find_gc_block(struct chfs_mount *chmp)
340 struct chfs_eraseblock *ret;
341 struct chfs_eraseblock_queue *nextqueue;
343 KASSERT(mutex_owned(&chmp->chm_lock_mountfields));
345 struct timespec now;
346 vfs_timestamp(&now);
348 int n = now.tv_nsec % 128;
350 //dbg_gc("n = %d\n", n);
351 again:
352 /* if (!TAILQ_EMPTY(&chmp->chm_bad_used_queue) && chmp->chm_nr_free_blocks > chmp->chm_nr_resv_blocks_gcbad) {
353 dbg_gc("Picking block from bad_used_queue to GC next\n");
354 nextqueue = &chmp->chm_bad_used_queue;
355 } else */if (n<50 && !TAILQ_EMPTY(&chmp->chm_erase_pending_queue)) {
356 dbg_gc("Picking block from erase_pending_queue to GC next\n");
357 nextqueue = &chmp->chm_erase_pending_queue;
358 } else if (n<110 && !TAILQ_EMPTY(&chmp->chm_very_dirty_queue) ) {
359 dbg_gc("Picking block from very_dirty_queue to GC next\n");
360 nextqueue = &chmp->chm_very_dirty_queue;
361 } else if (n<126 && !TAILQ_EMPTY(&chmp->chm_dirty_queue) ) {
362 dbg_gc("Picking block from dirty_queue to GC next\n");
363 nextqueue = &chmp->chm_dirty_queue;
364 } else if (!TAILQ_EMPTY(&chmp->chm_clean_queue)) {
365 dbg_gc("Picking block from clean_queue to GC next\n");
366 nextqueue = &chmp->chm_clean_queue;
367 } else if (!TAILQ_EMPTY(&chmp->chm_dirty_queue)) {
368 dbg_gc("Picking block from dirty_queue to GC next"
369 " (clean_queue was empty)\n");
370 nextqueue = &chmp->chm_dirty_queue;
371 } else if (!TAILQ_EMPTY(&chmp->chm_very_dirty_queue)) {
372 dbg_gc("Picking block from very_dirty_queue to GC next"
373 " (clean_queue and dirty_queue were empty)\n");
374 nextqueue = &chmp->chm_very_dirty_queue;
375 } else if (!TAILQ_EMPTY(&chmp->chm_erase_pending_queue)) {
376 dbg_gc("Picking block from erase_pending_queue to GC next"
377 " (clean_queue and {very_,}dirty_queue were empty)\n");
378 nextqueue = &chmp->chm_erase_pending_queue;
379 } else if (!TAILQ_EMPTY(&chmp->chm_erasable_pending_wbuf_queue)) {
380 dbg_gc("Synching wbuf in order to reuse "
381 "erasable_pendig_wbuf_queue blocks\n");
382 rw_enter(&chmp->chm_lock_wbuf, RW_WRITER);
383 chfs_flush_pending_wbuf(chmp);
384 rw_exit(&chmp->chm_lock_wbuf);
385 goto again;
386 } else {
387 dbg_gc("CHFS: no clean, dirty _or_ erasable"
388 " blocks to GC from! Where are they all?\n");
389 return NULL;
392 ret = TAILQ_FIRST(nextqueue);
393 if (chmp->chm_nextblock) {
394 dbg_gc("nextblock num: %u - gcblock num: %u\n",
395 chmp->chm_nextblock->lnr, ret->lnr);
396 if (ret == chmp->chm_nextblock)
397 goto again;
398 //KASSERT(ret != chmp->chm_nextblock);
399 //dbg_gc("first node lnr: %u ofs: %u\n", ret->first_node->lnr, ret->first_node->offset);
400 //dbg_gc("last node lnr: %u ofs: %u\n", ret->last_node->lnr, ret->last_node->offset);
402 TAILQ_REMOVE(nextqueue, ret, queue);
403 chmp->chm_gcblock = ret;
404 ret->gc_node = ret->first_node;
406 if (!ret->gc_node) {
407 dbg_gc("Oops! ret->gc_node at LEB: %u is NULL\n", ret->lnr);
408 panic("CHFS BUG - one LEB's gc_node is NULL\n");
411 /* TODO wasted size? */
412 return ret;
417 chfs_gcollect_pass(struct chfs_mount *chmp)
419 struct chfs_vnode_cache *vc;
420 struct chfs_eraseblock *eb;
421 struct chfs_node_ref *nref;
422 uint32_t gcblock_dirty;
423 struct chfs_inode *ip;
424 ino_t vno, pvno;
425 uint32_t nlink;
426 int ret = 0;
428 KASSERT(mutex_owned(&chmp->chm_lock_mountfields));
430 // mutex_enter(&chmp->chm_lock_mountfields);
431 for (;;) {
432 mutex_enter(&chmp->chm_lock_sizes);
434 dbg_gc("unchecked size == %u\n", chmp->chm_unchecked_size);
435 if (!chmp->chm_unchecked_size)
436 break;
438 if (chmp->chm_checked_vno > chmp->chm_max_vno) {
439 mutex_exit(&chmp->chm_lock_sizes);
440 mutex_exit(&chmp->chm_lock_mountfields);
441 dbg_gc("checked_vno (#%llu) > max_vno (#%llu)\n",
442 (unsigned long long)chmp->chm_checked_vno,
443 (unsigned long long)chmp->chm_max_vno);
444 return ENOSPC;
447 mutex_exit(&chmp->chm_lock_sizes);
449 mutex_enter(&chmp->chm_lock_vnocache);
450 dbg_gc("checking vno #%llu\n",
451 (unsigned long long)chmp->chm_checked_vno);
452 dbg_gc("get vnode cache\n");
453 vc = chfs_vnode_cache_get(chmp, chmp->chm_checked_vno++);
455 if (!vc) {
456 dbg_gc("!vc\n");
457 mutex_exit(&chmp->chm_lock_vnocache);
458 continue;
461 if ((vc->pvno | vc->nlink) == 0) {
462 dbg_gc("(pvno | nlink) == 0\n");
463 mutex_exit(&chmp->chm_lock_vnocache);
464 continue;
467 dbg_gc("switch\n");
468 switch (vc->state) {
469 case VNO_STATE_CHECKEDABSENT:
470 case VNO_STATE_PRESENT:
471 mutex_exit(&chmp->chm_lock_vnocache);
472 continue;
474 case VNO_STATE_GC:
475 case VNO_STATE_CHECKING:
476 mutex_exit(&chmp->chm_lock_vnocache);
477 mutex_exit(&chmp->chm_lock_mountfields);
478 dbg_gc("VNO_STATE GC or CHECKING\n");
479 panic("CHFS BUG - vc state gc or checking\n");
481 case VNO_STATE_READING:
482 chmp->chm_checked_vno--;
483 mutex_exit(&chmp->chm_lock_vnocache);
484 /* XXX why do we need the delay here?! */
485 kpause("chvncrea", true, mstohz(50), NULL);
487 // sleep_on_spinunlock(&chmp->chm_lock_vnocache);
488 // KASSERT(!mutex_owned(&chmp->chm_lock_vnocache));
489 mutex_exit(&chmp->chm_lock_mountfields);
490 return 0;
492 default:
493 mutex_exit(&chmp->chm_lock_vnocache);
494 mutex_exit(&chmp->chm_lock_mountfields);
495 dbg_gc("default\n");
496 panic("CHFS BUG - vc state is other what we"
497 " checked\n");
499 case VNO_STATE_UNCHECKED:
503 chfs_vnode_cache_set_state(chmp, vc, VNO_STATE_CHECKING);
505 /* XXX check if this is too heavy to call under
506 * chm_lock_vnocache
508 ret = chfs_check(chmp, vc);
509 dbg_gc("set state\n");
510 chfs_vnode_cache_set_state(chmp,
511 vc, VNO_STATE_CHECKEDABSENT);
513 mutex_exit(&chmp->chm_lock_vnocache);
514 mutex_exit(&chmp->chm_lock_mountfields);
516 return ret;
520 eb = chmp->chm_gcblock;
522 if (!eb) {
523 eb = find_gc_block(chmp);
526 if (!eb) {
527 dbg_gc("!eb\n");
528 if (!TAILQ_EMPTY(&chmp->chm_erase_pending_queue)) {
529 mutex_exit(&chmp->chm_lock_sizes);
530 mutex_exit(&chmp->chm_lock_mountfields);
531 return EAGAIN;
533 mutex_exit(&chmp->chm_lock_sizes);
534 mutex_exit(&chmp->chm_lock_mountfields);
535 return EIO;
538 if (!eb->used_size) {
539 dbg_gc("!eb->used_size\n");
540 goto eraseit;
543 nref = eb->gc_node;
544 //dbg_gc("gc use: %u\n", chmp->chm_nextblock->lnr);
545 //dbg_gc("nref: %u %u\n", nref->nref_lnr, nref->nref_offset);
546 gcblock_dirty = eb->dirty_size;
548 while(CHFS_REF_OBSOLETE(nref)) {
549 //dbg_gc("obsoleted nref lnr: %u - offset: %u\n", nref->nref_lnr, nref->nref_offset);
550 #ifdef DBG_MSG_GC
551 if (nref == chmp->chm_blocks[nref->nref_lnr].last_node) {
552 dbg_gc("THIS NODE IS THE LAST NODE OF ITS EB\n");
554 #endif
555 nref = node_next(nref);
556 if (!nref) {
557 //dbg_gc("!nref\n");
558 eb->gc_node = nref;
559 mutex_exit(&chmp->chm_lock_sizes);
560 mutex_exit(&chmp->chm_lock_mountfields);
561 panic("CHFS BUG - nref is NULL)\n");
564 eb->gc_node = nref;
565 //dbg_gc("nref the chosen one lnr: %u - offset: %u\n", nref->nref_lnr, nref->nref_offset);
566 KASSERT(nref->nref_lnr == chmp->chm_gcblock->lnr);
568 if (!nref->nref_next) {
569 //dbg_gc("!nref->nref_next\n");
570 mutex_exit(&chmp->chm_lock_sizes);
571 if (CHFS_REF_FLAGS(nref) == CHFS_PRISTINE_NODE_MASK) {
572 chfs_gcollect_pristine(chmp, eb, NULL, nref);
573 } else {
574 chfs_mark_node_obsolete(chmp, nref);
576 goto lock_size;
579 dbg_gc("nref lnr: %u - offset: %u\n", nref->nref_lnr, nref->nref_offset);
580 vc = chfs_nref_to_vc(nref);
582 mutex_exit(&chmp->chm_lock_sizes);
584 //dbg_gc("enter vnocache lock on #%llu\n", vc->vno);
585 mutex_enter(&chmp->chm_lock_vnocache);
587 dbg_gc("switch\n");
588 switch(vc->state) {
589 case VNO_STATE_CHECKEDABSENT:
590 if (CHFS_REF_FLAGS(nref) == CHFS_PRISTINE_NODE_MASK) {
591 chfs_vnode_cache_set_state(chmp, vc, VNO_STATE_GC);
593 break;
595 case VNO_STATE_PRESENT:
596 break;
598 case VNO_STATE_UNCHECKED:
599 case VNO_STATE_CHECKING:
600 case VNO_STATE_GC:
601 mutex_exit(&chmp->chm_lock_vnocache);
602 mutex_exit(&chmp->chm_lock_mountfields);
603 panic("CHFS BUG - vc state unchecked,"
604 " checking or gc (vno #%llu, num #%d)\n",
605 (unsigned long long)vc->vno, vc->state);
607 case VNO_STATE_READING:
608 mutex_exit(&chmp->chm_lock_vnocache);
609 /* XXX why do we need the delay here?! */
610 kpause("chvncrea", true, mstohz(50), NULL);
612 // sleep_on_spinunlock(&chmp->chm_lock_vnocache);
613 // KASSERT(!mutex_owned(&chmp->chm_lock_vnocache));
614 mutex_exit(&chmp->chm_lock_mountfields);
615 return 0;
618 if (vc->state == VNO_STATE_GC) {
619 dbg_gc("vc->state == VNO_STATE_GC\n");
620 mutex_exit(&chmp->chm_lock_vnocache);
621 ret = chfs_gcollect_pristine(chmp, eb, NULL, nref);
623 // chfs_vnode_cache_state_set(chmp,
624 // vc, VNO_STATE_CHECKEDABSENT);
625 /* XXX locking? */
626 vc->state = VNO_STATE_CHECKEDABSENT;
627 //TODO wake_up(&chmp->chm_vnocache_wq);
628 if (ret != EBADF)
629 goto test_gcnode;
630 mutex_enter(&chmp->chm_lock_vnocache);
633 vno = vc->vno;
634 pvno = vc->pvno;
635 nlink = vc->nlink;
636 mutex_exit(&chmp->chm_lock_vnocache);
638 ip = chfs_gc_fetch_inode(chmp, vno, !(pvno | nlink));
640 if (!ip) {
641 dbg_gc("!ip\n");
642 ret = 0;
643 goto lock_size;
646 chfs_gcollect_live(chmp, eb, nref, ip);
648 chfs_gc_release_inode(chmp, ip);
650 test_gcnode:
651 if (eb->dirty_size == gcblock_dirty &&
652 !CHFS_REF_OBSOLETE(eb->gc_node)) {
653 dbg_gc("ERROR collecting node at %u failed.\n",
654 CHFS_GET_OFS(eb->gc_node->nref_offset));
656 ret = ENOSPC;
659 lock_size:
660 KASSERT(mutex_owned(&chmp->chm_lock_mountfields));
661 mutex_enter(&chmp->chm_lock_sizes);
662 eraseit:
663 dbg_gc("eraseit\n");
665 if (chmp->chm_gcblock) {
666 dbg_gc("eb used size = %u\n", chmp->chm_gcblock->used_size);
667 dbg_gc("eb free size = %u\n", chmp->chm_gcblock->free_size);
668 dbg_gc("eb dirty size = %u\n", chmp->chm_gcblock->dirty_size);
669 dbg_gc("eb unchecked size = %u\n",
670 chmp->chm_gcblock->unchecked_size);
671 dbg_gc("eb wasted size = %u\n", chmp->chm_gcblock->wasted_size);
673 KASSERT(chmp->chm_gcblock->used_size + chmp->chm_gcblock->free_size +
674 chmp->chm_gcblock->dirty_size +
675 chmp->chm_gcblock->unchecked_size +
676 chmp->chm_gcblock->wasted_size == chmp->chm_ebh->eb_size);
680 if (chmp->chm_gcblock && chmp->chm_gcblock->dirty_size +
681 chmp->chm_gcblock->wasted_size == chmp->chm_ebh->eb_size) {
682 dbg_gc("Block at leb #%u completely obsoleted by GC, "
683 "Moving to erase_pending_queue\n", chmp->chm_gcblock->lnr);
684 TAILQ_INSERT_TAIL(&chmp->chm_erase_pending_queue,
685 chmp->chm_gcblock, queue);
686 chmp->chm_gcblock = NULL;
687 chmp->chm_nr_erasable_blocks++;
688 if (!TAILQ_EMPTY(&chmp->chm_erase_pending_queue)) {
689 ret = chfs_remap_leb(chmp);
693 mutex_exit(&chmp->chm_lock_sizes);
694 mutex_exit(&chmp->chm_lock_mountfields);
695 dbg_gc("return\n");
696 return ret;
701 chfs_gcollect_pristine(struct chfs_mount *chmp, struct chfs_eraseblock *cheb,
702 struct chfs_vnode_cache *chvc, struct chfs_node_ref *nref)
704 struct chfs_node_ref *newnref;
705 struct chfs_flash_node_hdr *nhdr;
706 struct chfs_flash_vnode *fvnode;
707 struct chfs_flash_dirent_node *fdirent;
708 struct chfs_flash_data_node *fdata;
709 int ret, retries = 0;
710 uint32_t ofs, crc;
711 size_t totlen = chfs_nref_len(chmp, cheb, nref);
712 char *data;
713 struct iovec vec;
714 size_t retlen;
716 dbg_gc("gcollect_pristine\n");
718 data = kmem_alloc(totlen, KM_SLEEP);
719 if (!data)
720 return ENOMEM;
722 ofs = CHFS_GET_OFS(nref->nref_offset);
724 ret = chfs_read_leb(chmp, nref->nref_lnr, data, ofs, totlen, &retlen);
725 if (ret) {
726 dbg_gc("reading error\n");
727 return ret;
729 if (retlen != totlen) {
730 dbg_gc("read size error\n");
731 return EIO;
733 nhdr = (struct chfs_flash_node_hdr *)data;
734 /* check the header */
735 if (le16toh(nhdr->magic) != CHFS_FS_MAGIC_BITMASK) {
736 dbg_gc("node header magic number error\n");
737 return EBADF;
739 crc = crc32(0, (uint8_t *)nhdr, CHFS_NODE_HDR_SIZE - 4);
740 if (crc != le32toh(nhdr->hdr_crc)) {
741 dbg_gc("node header crc error\n");
742 return EBADF;
745 switch(le16toh(nhdr->type)) {
746 case CHFS_NODETYPE_VNODE:
747 fvnode = (struct chfs_flash_vnode *)data;
748 crc = crc32(0, (uint8_t *)fvnode, sizeof(struct chfs_flash_vnode) - 4);
749 if (crc != le32toh(fvnode->node_crc)) {
750 dbg_gc("vnode crc error\n");
751 return EBADF;
753 break;
754 case CHFS_NODETYPE_DIRENT:
755 fdirent = (struct chfs_flash_dirent_node *)data;
756 crc = crc32(0, (uint8_t *)fdirent, sizeof(struct chfs_flash_dirent_node) - 4);
757 if (crc != le32toh(fdirent->node_crc)) {
758 dbg_gc("dirent crc error\n");
759 return EBADF;
761 crc = crc32(0, fdirent->name, fdirent->nsize);
762 if (crc != le32toh(fdirent->name_crc)) {
763 dbg_gc("dirent name crc error\n");
764 return EBADF;
766 break;
767 case CHFS_NODETYPE_DATA:
768 fdata = (struct chfs_flash_data_node *)data;
769 crc = crc32(0, (uint8_t *)fdata, sizeof(struct chfs_flash_data_node) - 4);
770 if (crc != le32toh(fdata->node_crc)) {
771 dbg_gc("data node crc error\n");
772 return EBADF;
774 break;
775 default:
776 if (chvc) {
777 dbg_gc("unknown node have vnode cache\n");
778 return EBADF;
781 /* CRC's OK, write node to its new place */
782 retry:
783 ret = chfs_reserve_space_gc(chmp, totlen);
784 if (ret)
785 return ret;
787 newnref = chfs_alloc_node_ref(chmp->chm_nextblock);
788 if (!newnref)
789 return ENOMEM;
791 ofs = chmp->chm_ebh->eb_size - chmp->chm_nextblock->free_size;
792 newnref->nref_offset = ofs;
794 vec.iov_base = (void *)data;
795 vec.iov_len = totlen;
796 mutex_enter(&chmp->chm_lock_sizes);
797 ret = chfs_write_wbuf(chmp, &vec, 1, ofs, &retlen);
799 if (ret || retlen != totlen) {
800 chfs_err("error while writing out to the media\n");
801 chfs_err("err: %d | size: %zu | retlen : %zu\n",
802 ret, totlen, retlen);
804 chfs_change_size_dirty(chmp, chmp->chm_nextblock, totlen);
805 if (retries) {
806 mutex_exit(&chmp->chm_lock_sizes);
807 return EIO;
810 retries++;
811 mutex_exit(&chmp->chm_lock_sizes);
812 goto retry;
815 mutex_exit(&chmp->chm_lock_sizes);
816 //TODO should we set free_size?
817 chfs_mark_node_obsolete(chmp, nref);
818 chfs_add_vnode_ref_to_vc(chmp, chvc, newnref);
819 return 0;
824 chfs_gcollect_live(struct chfs_mount *chmp,
825 struct chfs_eraseblock *cheb, struct chfs_node_ref *nref,
826 struct chfs_inode *ip)
828 struct chfs_node_frag *frag;
829 struct chfs_full_dnode *fn = NULL;
830 int start = 0, end = 0, nrfrags = 0;
831 struct chfs_dirent *fd = NULL;
832 int ret = 0;
833 bool is_dirent;
835 dbg_gc("gcollect_live\n");
837 if (chmp->chm_gcblock != cheb) {
838 dbg_gc("GC block is no longer gcblock. Restart.\n");
839 goto upnout;
842 if (CHFS_REF_OBSOLETE(nref)) {
843 dbg_gc("node to be GC'd was obsoleted in the meantime.\n");
844 goto upnout;
847 /* It's a vnode? */
848 if (ip->chvc->v == nref) {
849 chfs_gcollect_vnode(chmp, ip);
850 goto upnout;
853 /* find fn */
854 dbg_gc("find full dnode\n");
855 for(frag = frag_first(&ip->fragtree);
856 frag; frag = frag_next(&ip->fragtree, frag)) {
857 if (frag->node && frag->node->nref == nref) {
858 fn = frag->node;
859 end = frag->ofs + frag->size;
860 if (!nrfrags++)
861 start = frag->ofs;
862 if (nrfrags == frag->node->frags)
863 break;
867 /* It's a pristine node, or dnode (or hole? XXX have we hole nodes?) */
868 if (fn) {
869 if (CHFS_REF_FLAGS(nref) == CHFS_PRISTINE_NODE_MASK) {
870 ret = chfs_gcollect_pristine(chmp,
871 cheb, ip->chvc, nref);
872 if (!ret) {
873 frag->node->nref = ip->chvc->v;
875 if (ret != EBADF)
876 goto upnout;
878 //ret = chfs_gcollect_hole(chmp, cheb, ip, fn, start, end);
879 ret = chfs_gcollect_dnode(chmp, cheb, ip, fn, start, end);
880 goto upnout;
884 /* It's a dirent? */
885 dbg_gc("find full dirent\n");
886 is_dirent = false;
887 TAILQ_FOREACH(fd, &ip->dents, fds) {
888 if (fd->nref == nref) {
889 is_dirent = true;
890 break;
894 if (is_dirent && fd->vno) {
895 ret = chfs_gcollect_dirent(chmp, cheb, ip, fd);
896 } else if (is_dirent) {
897 ret = chfs_gcollect_deletion_dirent(chmp, cheb, ip, fd);
898 } else {
899 dbg_gc("Nref at leb #%u offset 0x%08x wasn't in node list"
900 " for ino #%llu\n",
901 nref->nref_lnr, CHFS_GET_OFS(nref->nref_offset),
902 (unsigned long long)ip->ino);
903 if (CHFS_REF_OBSOLETE(nref)) {
904 dbg_gc("But it's obsolete so we don't mind"
905 " too much.\n");
909 upnout:
910 return ret;
914 chfs_gcollect_vnode(struct chfs_mount *chmp, struct chfs_inode *ip)
916 int ret;
917 dbg_gc("gcollect_vnode\n");
919 ret = chfs_write_flash_vnode(chmp, ip, ALLOC_GC);
921 return ret;
925 chfs_gcollect_dirent(struct chfs_mount *chmp,
926 struct chfs_eraseblock *cheb, struct chfs_inode *parent,
927 struct chfs_dirent *fd)
929 struct vnode *vnode = NULL;
930 struct chfs_inode *ip;
931 struct chfs_node_ref *prev;
932 dbg_gc("gcollect_dirent\n");
934 vnode = chfs_vnode_lookup(chmp, fd->vno);
936 /* XXX maybe KASSERT or panic on this? */
937 if (vnode == NULL) {
938 return ENOENT;
941 ip = VTOI(vnode);
943 prev = parent->chvc->dirents;
944 if (prev == fd->nref) {
945 parent->chvc->dirents = prev->nref_next;
946 dbg_gc("fd nref removed from dirents list\n");
947 prev = NULL;
949 while (prev) {
950 if (prev->nref_next == fd->nref) {
951 prev->nref_next = fd->nref->nref_next;
952 dbg_gc("fd nref removed from dirents list\n");
953 break;
955 prev = prev->nref_next;
958 prev = fd->nref;
959 chfs_mark_node_obsolete(chmp, fd->nref);
960 return chfs_write_flash_dirent(chmp,
961 parent, ip, fd, fd->vno, ALLOC_GC);
964 /* Check dirents what are marked as deleted. */
966 chfs_gcollect_deletion_dirent(struct chfs_mount *chmp,
967 struct chfs_eraseblock *cheb, struct chfs_inode *parent,
968 struct chfs_dirent *fd)
970 struct chfs_flash_dirent_node chfdn;
971 struct chfs_node_ref *nref;
972 size_t retlen, name_len, nref_len;
973 uint32_t name_crc;
975 int ret;
977 struct vnode *vnode = NULL;
979 dbg_gc("gcollect_deletion_dirent\n");
981 name_len = strlen(fd->name);
982 name_crc = crc32(0, fd->name, name_len);
984 nref_len = chfs_nref_len(chmp, cheb, fd->nref);
986 vnode = chfs_vnode_lookup(chmp, fd->vno);
988 //dbg_gc("ip from vnode\n");
989 //VFS_VGET(chmp->chm_fsmp, fd->vno, &vnode);
990 //ip = VTOI(vnode);
991 //vput(vnode);
993 //dbg_gc("mutex enter erase_completion_lock\n");
995 // dbg_gc("alloc chfdn\n");
996 // chfdn = kmem_alloc(nref_len, KM_SLEEP);
997 // if (!chfdn)
998 // return ENOMEM;
1000 for (nref = parent->chvc->dirents;
1001 nref != (void*)parent->chvc;
1002 nref = nref->nref_next) {
1004 if (!CHFS_REF_OBSOLETE(nref))
1005 continue;
1007 /* if node refs have different length, skip */
1008 if (chfs_nref_len(chmp, NULL, nref) != nref_len)
1009 continue;
1011 if (CHFS_GET_OFS(nref->nref_offset) ==
1012 CHFS_GET_OFS(fd->nref->nref_offset)) {
1013 continue;
1016 ret = chfs_read_leb(chmp,
1017 nref->nref_lnr, (void*)&chfdn, CHFS_GET_OFS(nref->nref_offset),
1018 nref_len, &retlen);
1020 if (ret) {
1021 dbg_gc("Read error: %d\n", ret);
1022 continue;
1025 if (retlen != nref_len) {
1026 dbg_gc("Error reading node:"
1027 " read: %zu insted of: %zu\n", retlen, nref_len);
1028 continue;
1031 /* if node type doesn't match, skip */
1032 if (le16toh(chfdn.type) != CHFS_NODETYPE_DIRENT)
1033 continue;
1035 /* if crc doesn't match, skip */
1036 if (le32toh(chfdn.name_crc) != name_crc)
1037 continue;
1039 /* if length of name different, or this is an another deletion
1040 * dirent, skip
1042 if (chfdn.nsize != name_len || !le64toh(chfdn.vno))
1043 continue;
1045 /* check actual name */
1046 if (memcmp(chfdn.name, fd->name, name_len))
1047 continue;
1049 // kmem_free(chfdn, nref_len);
1051 chfs_mark_node_obsolete(chmp, fd->nref);
1052 return chfs_write_flash_dirent(chmp,
1053 parent, NULL, fd, fd->vno, ALLOC_GC);
1056 // kmem_free(chfdn, nref_len);
1058 TAILQ_REMOVE(&parent->dents, fd, fds);
1059 chfs_free_dirent(fd);
1060 return 0;
1064 chfs_gcollect_dnode(struct chfs_mount *chmp,
1065 struct chfs_eraseblock *orig_cheb, struct chfs_inode *ip,
1066 struct chfs_full_dnode *fn, uint32_t orig_start, uint32_t orig_end)
1068 struct chfs_node_ref *nref, *prev;
1069 struct chfs_full_dnode *newfn;
1070 struct chfs_flash_data_node *fdnode;
1071 int ret = 0, retries = 0;
1072 uint32_t totlen;
1073 char *data = NULL;
1074 struct iovec vec;
1075 size_t retlen;
1076 dbg_gc("gcollect_dnode\n");
1078 //uint32_t used_size;
1080 /* TODO GC merging frags, should we use it?
1082 uint32_t start, end;
1084 start = orig_start;
1085 end = orig_end;
1087 if (chmp->chm_nr_free_blocks + chmp->chm_nr_erasable_blocks > chmp->chm_resv_blocks_gcmerge) {
1088 struct chfs_node_frag *frag;
1089 uint32_t min, max;
1091 min = start & (PAGE_CACHE_SIZE-1);
1092 max = min + PAGE_CACHE_SIZE;
1094 frag = (struct chfs_node_frag *)rb_tree_find_node_leq(&ip->i_chfs_ext.fragtree, &start);
1095 KASSERT(frag->ofs == start);
1097 while ((frag = frag_prev(&ip->i_chfs_ext.fragtree, frag)) && frag->ofs >= min) {
1098 if (frag->ofs > min) {
1099 start = frag->ofs;
1100 continue;
1103 if (!frag->node || !frag->node->nref) {
1104 break;
1105 } else {
1106 struct chfs_node_ref *nref = frag->node->nref;
1107 struct chfs_eraseblock *cheb;
1109 cheb = &chmp->chm_blocks[nref->nref_lnr];
1111 if (cheb == chmp->chm_gcblock)
1112 start = frag->ofs;
1114 //TODO is this a clean block?
1116 start = frag->ofs;
1117 break;
1121 end--;
1122 frag = (struct chfs_node_frag *)rb_tree_find_node_leq(&ip->i_chfs_ext.fragtree, &(end));
1124 while ((frag = frag_next(&ip->i_chfs_ext.fragtree, frag)) && (frag->ofs + frag->size <= max)) {
1125 if (frag->ofs + frag->size < max) {
1126 end = frag->ofs + frag->size;
1127 continue;
1130 if (!frag->node || !frag->node->nref) {
1131 break;
1132 } else {
1133 struct chfs_node_ref *nref = frag->node->nref;
1134 struct chfs_eraseblock *cheb;
1136 cheb = &chmp->chm_blocks[nref->nref_lnr];
1138 if (cheb == chmp->chm_gcblock)
1139 end = frag->ofs + frag->size;
1141 //TODO is this a clean block?
1143 end = frag->ofs + frag->size;
1144 break;
1148 KASSERT(end <=
1149 frag_last(&ip->i_chfs_ext.fragtree)->ofs +
1150 frag_last(&ip->i_chfs_ext.fragtree)->size);
1151 KASSERT(end >= orig_end);
1152 KASSERT(start <= orig_start);
1155 KASSERT(orig_cheb->lnr == fn->nref->nref_lnr);
1156 totlen = chfs_nref_len(chmp, orig_cheb, fn->nref);
1157 data = kmem_alloc(totlen, KM_SLEEP);
1159 ret = chfs_read_leb(chmp, fn->nref->nref_lnr, data, fn->nref->nref_offset,
1160 totlen, &retlen);
1162 fdnode = (struct chfs_flash_data_node *)data;
1163 fdnode->version = htole64(++ip->chvc->highest_version);
1164 fdnode->node_crc = htole32(crc32(0, (uint8_t *)fdnode,
1165 sizeof(*fdnode) - 4));
1167 vec.iov_base = (void *)data;
1168 vec.iov_len = totlen;
1170 retry:
1171 ret = chfs_reserve_space_gc(chmp, totlen);
1172 if (ret)
1173 goto out;
1175 nref = chfs_alloc_node_ref(chmp->chm_nextblock);
1176 if (!nref) {
1177 ret = ENOMEM;
1178 goto out;
1181 mutex_enter(&chmp->chm_lock_sizes);
1183 nref->nref_offset = chmp->chm_ebh->eb_size - chmp->chm_nextblock->free_size;
1184 KASSERT(nref->nref_offset % 4 == 0);
1185 chfs_change_size_free(chmp, chmp->chm_nextblock, -totlen);
1187 ret = chfs_write_wbuf(chmp, &vec, 1, nref->nref_offset, &retlen);
1188 if (ret || retlen != totlen) {
1189 chfs_err("error while writing out to the media\n");
1190 chfs_err("err: %d | size: %d | retlen : %zu\n",
1191 ret, totlen, retlen);
1192 chfs_change_size_dirty(chmp, chmp->chm_nextblock, totlen);
1193 if (retries) {
1194 ret = EIO;
1195 mutex_exit(&chmp->chm_lock_sizes);
1196 goto out;
1199 retries++;
1200 mutex_exit(&chmp->chm_lock_sizes);
1201 goto retry;
1204 dbg_gc("new nref lnr: %u - offset: %u\n", nref->nref_lnr, nref->nref_offset);
1206 chfs_change_size_used(chmp, &chmp->chm_blocks[nref->nref_lnr], totlen);
1207 mutex_exit(&chmp->chm_lock_sizes);
1208 KASSERT(chmp->chm_blocks[nref->nref_lnr].used_size <= chmp->chm_ebh->eb_size);
1210 newfn = chfs_alloc_full_dnode();
1211 newfn->nref = nref;
1212 newfn->ofs = fn->ofs;
1213 newfn->size = fn->size;
1214 newfn->frags = fn->frags;
1216 //TODO should we remove fd from dnode list?
1218 prev = ip->chvc->dnode;
1219 if (prev == fn->nref) {
1220 ip->chvc->dnode = prev->nref_next;
1221 prev = NULL;
1223 while (prev) {
1224 if (prev->nref_next == fn->nref) {
1225 prev->nref_next = fn->nref->nref_next;
1226 break;
1228 prev = prev->nref_next;
1231 chfs_add_full_dnode_to_inode(chmp, ip, newfn);
1232 chfs_add_node_to_list(chmp,
1233 ip->chvc, newfn->nref, &ip->chvc->dnode);
1235 out:
1236 kmem_free(data, totlen);
1237 return ret;